Detection of Atrial Fibrillation in a Large Population Using Wearable Devices: The Fitbit Heart Study
Comprehensive Overview
Published in Circulation in 2022, this study asked whether a smartwatch algorithm could reliably flag undiagnosed atrial fibrillation, an irregular heart rhythm that raises the risk of stroke and often goes unnoticed because it can be silent, while keeping false alarms to a minimum. Researchers conducted a prospective remote clinical trial, the Fitbit Heart Study, enrolling 455,699 US adults aged 22 or older who used a compatible Fitbit device and had no prior diagnosis of atrial fibrillation. A novel software algorithm continuously analyzed each participant's optical pulse-sensor data during periods of stillness, such as sleep, and flagged an irregular heart rhythm detection whenever 11 consecutive overlapping five-minute pulse readings appeared irregular. Participants who received this notification were invited to a telehealth visit with a study physician and mailed a one-week ambulatory ECG patch monitor to confirm, through independent cardiologist review, whether atrial fibrillation was actually present. The primary outcome was the positive predictive value of the algorithm's first irregular rhythm detection for atrial fibrillation confirmed on that patch monitor. The topline result was that the algorithm's detections corresponded to confirmed atrial fibrillation with very high accuracy among those who returned analyzable ECG monitoring data.
Outcomes and Implications
The most important finding was that among the 1,057 participants who received an irregular heart rhythm notification and returned an analyzable ECG patch, the algorithm's first detection had a positive predictive value of 98.2% for concurrent atrial fibrillation, meaning that when the device flagged a second irregular rhythm during patch monitoring, it was almost always confirmed as true atrial fibrillation on the simultaneous ECG. This accuracy held up well across subgroups, including a 97.0% positive predictive value among participants aged 65 and older, who face the greatest stroke risk from undiagnosed atrial fibrillation. Only about 1% of all enrolled participants ever received an irregular rhythm notification, and of those who completed ECG patch monitoring, roughly one in three had confirmed atrial fibrillation, with a median burden of 7% of monitored time and a median longest episode lasting 7 hours. Most first detections occurred during sleep, when the device could most reliably sense a stable pulse. The algorithm's sensitivity for catching atrial fibrillation episodes during the monitoring week was more modest at 67.6%, reflecting that fibrillation can be intermittent and easy to miss with any single detection window. The authors conclude that this wearable-based algorithm reliably identifies people likely to have undiagnosed atrial fibrillation with a high degree of confidence, suggesting that widely available consumer devices could meaningfully expand opportunistic screening, though prospective studies are still needed to establish how such detections should translate into clinical care and stroke prevention.